Communications of the ACM
ML Models for Diagnosing Covid-19 Not Yet Suitable for Clinical Use
A review of studies suggesting machine learning models for detecting and diagnosing COVID-19 from medical images has found major flaws and biases in those models.
Credit: NIAID
A review of studies containing descriptions of machine learning (ML) models for diagnosing Covid-19 by researchers at the U.K.'s University of Cambridge concluded that none are yet suitable for detecting or diagnosing the virus from standard medical imaging.
The team ultimately reviewed 62 studies, and invalidated each model's suitability due to biases in study design, methodological flaws, lack of reproducibility, and publicly available "Frankenstein datasets."
Many ML models were trained on sample datasets that were too small to be effective, failed to specify their data's origins, were trained and tested on the same data, or lacked involvement from radiologists and clinicians.
Cambridge's James Rudd said, "These early studies show promise, but they suffer from a high prevalence of deficiencies in methodology and reporting, with none of the literature we reviewed reaching the threshold of robustness and reproducibility essential to support use in clinical practice."
From University of Cambridge (U.K.)
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found